1. Data privacy and security in the world of large language models
Deepak Kumar
AI Principal Technologist - Windriver
https://www.linkedin.com/in/dpkumar/
2. Agenda
● Cybersecurity challenges
● Data breaches/leaks in the recent past
● Data privacy with ChatGPT's policy to retain data
● Brainstorming
6. Issues
● AI-Generated Phishing Scams
○ Threat actors to use ChatGPT to create sophisticated and realistic
phishing emails.
○ A similarly dangerous tactic is the use of AI to create disinformation and
conspiracy campaigns.
● Duping ChatGPT into Writing Malicious Code
○ The chatbot can be prompted to generate malicious code
Reference: https://hbr.org/2023/04/the-new-risks-chatgpt-poses-to-cybersecurity
7. Issues
● Increase in attack surface due to chatGPT privacy policy
○ it may collect personal information from your messages, any files you
upload
● Human engineering based attack
○ Meta said it had “investigated and taken action against malware strains
taking advantage of people’s interest in OpenAI’s ChatGPT to trick them
into installing malware pretending to provide AI functionality.”
Reference: https://edition.cnn.com/2023/05/03/tech/chatgpt-hackers-meta/index.html
16. Privacy problems
● ChatGPT’s privacy policy issue
○ Collect personal information from your messages. It
includes the document you upload to chatGPT
○ Your conversations may be reviewed by its AI trainers
to improve the chat and train the system further
● Not complying the PII policies like GDPR
18. OpenAI Handling
● Bug bounty of up to $20,000 to anyone who discovers unreported vulnerabilities.
Reference: https://venturebeat.com/security/privateais-privategpt-aims-to-combat-chatgpt-privacy-concerns/
https://www.businesstoday.in/technology/news/story/samsung-employees-accidentally-leaked-company-secrets-via-
chatgpt-heres-what-happened-376375-2023-04-06
https://cybernews.com/news/chatgpt-samsung-data-leak/